/*

Here's a way to solve a recurrent problem : emulating glLineWidth in Direct3D. Better : you can use this code to draw textured lines, which are often needed to create a lot of special effects (lightning bolts, toon-like silhouettes, etc).
 
The code computes a world quad from a segment in world space. Simply render the quad instead of the segment! If you have a line-strip, compute one quad for each segment, then render everything as a single non-indexed tristrip of 4 * NbSegments vertices.
 
You should be able to use it with any flexible vertex format (FVF).
 
Here's the main routine :

*/
 
//////////////////////////////////////////////////////////////////////////////////////////////////////////
/**
 * Computes a screen quad aligned with a world segment.
 * \param  inverseview  [in] inverse view matrix
 * \param  view   [in] view matrix
 * \param  proj   [in] projection matrix
 * \param  verts   [out] 4 quad vertices in world space (forming a tri-strip)
 * \param  uvs    [out] 4 quad uvs
 * \param  stride   [in] size of vertex (FVF stride)
 * \param  p0    [in] segment's first point in world space
 * \param  p1    [in] segment's second point in world space
 * \param  size   [in] size of segment/quad
 * \param  constantsize [in] true to keep the quad's screen size constant (e.g. needed to emulate glLineWidth)
 */
//////////////////////////////////////////////////////////////////////////////////////////////////////////
void Renderer::ComputeScreenQuad(const ViewMatrix& inverseview, const ViewMatrix& view, const ProjMatrix& proj, ubyte* verts, ubyte* uvs, udword stride, const Point& p0, const Point& p1, float size, bool constantsize)
{
 // Compute delta in camera space
 Point Delta; TransformPoint3x3(Delta, p1-p0, view);
 
 // Compute size factors
 float SizeP0 = size;
 float SizeP1 = size;
 
 if(constantsize)
 {
  // Compute scales so that screen-size is constant
  SizeP0 *= ComputeConstantScale(p0, view, proj);
  SizeP1 *= ComputeConstantScale(p1, view, proj);
 }
 
 // Compute quad vertices
 float Theta0 = atan2f(-Delta.x, -Delta.y);
 float c0 = SizeP0 * cosf(Theta0);
 float s0 = SizeP0 * sinf(Theta0);
 ComputePoint(*((Point*)verts),  c0, -s0, inverseview, p0); verts+=stride;
 ComputePoint(*((Point*)verts),  -c0, s0, inverseview, p0); verts+=stride;
 
 float Theta1 = atan2f(Delta.x, Delta.y);
 float c1 = SizeP1 * cosf(Theta1);
 float s1 = SizeP1 * sinf(Theta1);
 ComputePoint(*((Point*)verts),  -c1, s1, inverseview, p1); verts+=stride;
 ComputePoint(*((Point*)verts),  c1, -s1, inverseview, p1); verts+=stride;
 
 // Output uvs if needed
 if(uvs)
 {
  *((float*)uvs) = 0.0f; *((float*)(uvs+4)) = 1.0f; uvs+=stride;
  *((float*)uvs) = 0.0f; *((float*)(uvs+4)) = 0.0f; uvs+=stride;
  *((float*)uvs) = 1.0f; *((float*)(uvs+4)) = 1.0f; uvs+=stride;
  *((float*)uvs) = 1.0f; *((float*)(uvs+4)) = 0.0f; uvs+=stride;
 }
}

inline void ComputePoint(Point& dest, float x, float y, const Matrix4x4& rot, const Point& trans)
{
 dest.x = trans.x + x * rot.m[0][0] + y * rot.m[1][0];
 dest.y = trans.y + x * rot.m[0][1] + y * rot.m[1][1];
 dest.z = trans.z + x * rot.m[0][2] + y * rot.m[1][2];
}

 
And extra used functions:
 
// Quickly rotates a vector, using the 3x3 part of a 4x4 matrix
 inline void TransformPoint3x3(Point& dest, const Point& source, const Matrix4x4& rot)
 {
  dest.x = source.x * rot.m[0][0] + source.y * rot.m[1][0] + source.z * rot.m[2][0];
  dest.y = source.x * rot.m[0][1] + source.y * rot.m[1][1] + source.z * rot.m[2][1];
  dest.z = source.x * rot.m[0][2] + source.y * rot.m[1][2] + source.z * rot.m[2][2];
 }

float Renderer::ComputeConstantScale(const Point& pos, const ViewMatrix& view, const ProjMatrix& proj)
{
 Point ppcam0 = pos * view;
 Point ppcam1 = ppcam0;
 ppcam1.x += 1.0f;
 
 float l1 = 1.0f/(ppcam0.x*proj.m[0][3] + ppcam0.y*proj.m[1][3] + ppcam0.z*proj.m[2][3] + proj.m[3][3]);
 float c1 =  (ppcam0.x*proj.m[0][0] + ppcam0.y*proj.m[1][0] + ppcam0.z*proj.m[2][0] + proj.m[3][0])*l1;
 float l2 = 1.0f/(ppcam1.x*proj.m[0][3] + ppcam1.y*proj.m[1][3] + ppcam1.z*proj.m[2][3] + proj.m[3][3]);
 float c2 =  (ppcam1.x*proj.m[0][0] + ppcam1.y*proj.m[1][0] + ppcam1.z*proj.m[2][0] + proj.m[3][0])*l2;
 float CorrectScale = 1.0f/(c2 - c1);
 return CorrectScale / float(mRenderWidth);
}
 
 
All matrices are 4x4 D3D-compliant matrices. mRenderWidth is the width of the render window (ex: 640)
 
I hope it will be useful to some of you. I know I would have loved to have this right from the start (read: included in D3D...). By the way, it's not supposed to be optimized so don't bother telling me it's not :)
 
 
Pierre